基于单幅立式标靶图像的单目深度信息提取  

Depth information extraction of on-board monocular vision based on a single vertical target image

作　　者：黄小云[1] 高峰[1] 徐国艳[1] 丁能根[1] 邢龙龙[1] 

机构地区：[1]北京航空航天大学交通科学与工程学院,北京100191

出　　处：《北京航空航天大学学报》2015年第4期649-655,共7页Journal of Beijing University of Aeronautics and Astronautics

基　　金：国家自然科学基金资助项目(51105021);国家自然科学基金青年基金资助项目(51405008);北京市自然科学基金资助项目(3133040)

摘　　要：针对在智能车中车载单目视觉系统已经检测到路面障碍物的情况下,计算障碍物相对于本车的距离问题,提出了一种仅利用单幅标靶图像且无需相机内部参数的图像深度信息提取方法.该方法利用一种放置于相机前方的立式标靶,建立图像纵坐标像素与实际成像角度之间的映射关系,结合投影几何模型实现实时深度信息的提取.依据立式标靶图像的特点,设计了包括标靶图像感兴趣区设置、模板匹配、候选点聚类、筛选及精确定位等处理的亚像素级角点检测及定位算法.实验结果显示,该方法具有较高的测量精度及实时性.相对于在路面摆放参照物的方法,该方法无需大标定场地且规避了数据拟合引起的误差.同时该方法标定只需一幅图像,过程简单,便于实际应用.Calculating obstacle distance for on-board monocular vision system on intelligent vehicle was investigated when the obstacle has been detected. An extraction method of depth information only using a single target image without any internal camera parameters was developed. The mapping relation between image row pixel values and the actual imaging angles was established with the image of vertical target, which was placed in the front of a camera. The obstacle depth information was extracted in real time by combining the projection geometry model. Given the characteristic of vertical target image, an algorithm of sub-pixel corner detection and location was designed, which includes region of interest setting, template matching, candidate points clustering and screening and precise location, etc. Experimental results show that the method has high precision and real-time performance. Compared with the method of putting reference on the road, it does not need large calibration site and could avoid the data fitting error. And the method also has a simple calibration procedure with a single image, which is suitable for practical application.

关 键 词：机器视觉 单目测距 立式标靶 智能车 角点检测 

分 类 号：U491[交通运输工程—交通运输规划与管理]